Researchers created a fictional eye condition called “bixonimania” to test whether AI systems could distinguish legitimate medical information from fabricated content. The fake condition was intentionally filled with obvious warning signs, including fictional authors, nonexistent institutions, and references that should have made it clear the material was not genuine. Despite these clues, several AI chatbots began describing the condition as if it were a real medical disorder.
The experiment was led by researcher Almira Osmanovic Thunström, who uploaded fabricated studies and supporting content online. Within weeks, major AI systems were repeating details about the nonexistent condition, providing explanations of symptoms, causes, and potential treatments. The researchers found that the models often prioritized patterns that looked scientifically credible rather than verifying whether the underlying information was authentic.
Perhaps more concerning, the misinformation did not remain confined to chatbots. The fake studies were reportedly cited in real academic work before being identified and withdrawn, suggesting that some researchers may have relied on AI-generated references or failed to verify original sources. The incident demonstrated how fabricated information can move from obscure online content into broader information ecosystems when both humans and AI systems neglect rigorous fact-checking.
The case highlights a broader challenge facing artificial intelligence: models can generate convincing answers without understanding whether the information is true. Researchers say the findings underscore the need for stronger verification mechanisms, better source evaluation, and greater user skepticism when seeking medical or scientific advice from AI systems. The experiment serves as a reminder that confidence and accuracy are not the same thing, especially in high-stakes fields such as healthcare.
